Physical Review Applied, 25, 3, 034072 (2026)
Reconfigurable photonics have rapidly become an invaluable tool for information processing. Light-based computing accelerators are promising for boosting neural-network learning and inference [X. Xiao et al., APL Photonics 6, 126107 (2021); X. Meng et al., Nat. Commun. 14, 3000 (2023); J. Cheng et al., Nat. Commun. 15, 6189 (2024)] and optical interconnects are foreseen as a solution to the information transfer bottleneck in high-performance computing [Y. Li et al., in 2021 58th ACM/IEEE Design Automation Conference (DAC) (IEEE, 2021), pp. 931–936; A. Rizzo et al., IEEE J. Sel. Top. Quantum Electron. 29, 1 (2022); A. Netherton et al., Photonics Res. 12, A69 (2024)]. In this study, we demonstrate the successful programming of a transformation implemented using a reconfigurable photonic circuit with a nonconventional architecture. The core of most photonic processors is an MZI-based architecture [M. Reck et al., Phys. Rev. Lett. 73, 58 (1994); W. R. Clements et al., Optica 3, 1460 (2016)] that establishes an analytical connection between controllable parameters and circuit transformation. However, several architectures [R. Tang et al., IEEE Photonics Technol. Lett. 29, 971 (2017); M. Y. Saygin et al., Phys. Rev. Lett. 124, 010501 (2020)] that are substantially more difficult to program have improved robustness to fabrication defects. We use two algorithms that rely on different initial datasets to reconstruct the circuit model of a complex interferometer, and then program the required unitary transformation. The first method is based on the global fitting of the experimental calibration data, while the second method is an ML-based approach introduced in Kuzmin et al. [Opt. Express 29, 38429 (2021)]. Both methods performed accurate circuit programming with an average fidelity greater than 99% and 97%, respectively. Our results provide a strong foundation for the introduction of nonconventional interferometric architectures for photonic information processing.